Windows supports examples 01–10. Examples 11–17 and the supplied SAM3D, robot and RoMa setups require Linux. Default Windows preparation skips RoMa; explicit requests report the platform restriction. GroundingDINO source preparation requires Git on PATH: install Git, reopen the terminal and verify git --version.

Windows 支持示例 01–10。示例 11–17 及当前 SAM3D、机器人、RoMa 安装方案需要 Linux。Windows 默认准备跳过 RoMa；显式请求时提示平台限制。准备 GroundingDINO 源码需要 PATH 中可用的 Git：请安装 Git，重开终端并用 git --version 确认。

Extension setup扩展环境

Start with the installation guide. Export bundled examples using export_examples; source-checkout users can run the repository's examples/ directly.

先按安装引导准备环境。随包示例可通过 export_examples 导出；源码用户可直接运行仓库中的 examples/。

Optional wheel featureswheel 可选功能

Basic installation does not install all optional frameworks. Prepare only the feature you need, or let its supported first-use entry prepare it. Extras such as wapr[det2d] provide helper dependencies; they do not replace feature preparation. Examples 01, 08, 09 and 12 prepare detection dependencies before use; importing these examples alone does not install optional features. Ultralytics is not a core pip dependency. If preparation is declined or incompatible, the feature stops before importing its optional framework. Inspect compatibility before installation:

基础安装不会安装全部可选框架。只准备需要的功能，或由支持首次使用准备的入口触发。wapr[det2d] 等 extras 提供辅助依赖，不代替功能准备。示例 01、08、09、12 会先准备检测依赖再使用，仅导入这些示例不会启动可选安装；Ultralytics 不属于 pip 基础依赖。准备被拒绝或环境不兼容时，对应功能会在导入可选框架前停止。可先检查兼容性：

--check inspects local dependencies without downloading or resolving an installation plan. Run preparation without --check to install missing dependencies; proposed replacements require confirmation.

--check 只检查本地依赖，不下载或联网解析安装计划。去掉 --check 可准备缺失依赖；更换已有依赖前会请求确认。

python -m wapr.bootstrap --feature det2d --check
python -m wapr.bootstrap --feature det2d
python -m wapr.bootstrap --feature sam3d
python -m wapr.bootstrap --feature robot

SAM3D uses an independent Python 3.11 environment under cache/environments/sam3d/. Managed Python interpreters are stored under cache/interpreters/; pose estimation retains its existing Python and PyTorch environment. SAM3D weights download from ModelScope by default, or reuse existing local weights. The upstream model license still applies. Examples 11 and 12 both default to cracker; mustard images elsewhere show recorded examples.

SAM3D 使用 cache/environments/sam3d/ 下的独立 Python 3.11 环境，托管解释器放在 cache/interpreters/；位姿估计保留原有 Python 与 PyTorch 环境。SAM3D 权重默认从魔搭下载，也可复用已有本地权重；使用仍须遵守上游模型许可。示例 11、12 默认均为 cracker；其他页面的 mustard 图片属于已录制展示。

Use --feature compatible to attempt the optional features supported by the current environment. Unsupported features are reported. Changes to existing dependencies require approval; inspect the proposed changes before accepting.

使用 --feature compatible 可尝试准备当前环境支持的可选功能；不支持的功能会报告原因。更换已有依赖需要确认，请先查看变更内容。

After installing the detection and pose environment, prepare only the extra dependencies needed by the application you use.

完成检测与位姿环境安装后，再按需准备以下应用的额外依赖。

Reconstruction物体重建

Choose reconstruction and robot extensions for the examples you want to run. SAM3D uses a compatible Linux environment and downloads weights from ModelScope by default. Upstream recommends about 32 GB GPU memory; actual requirements depend on the selected configuration.

根据需要运行的示例，准备重建或机器人扩展。SAM3D 使用兼容的 Linux 独立环境，默认从魔搭下载权重，上游建议约 32 GB 显存，实际容量以所选配置为准。

UniPose9D^[5] and Qwen^[7] are prepared when requested. Reconstruction resources use the same WAPR_CACHE_DIR root as pose estimation: source dependencies in cache/sources/, model files in cache/weights/, and shared demonstration inputs in cache/samples/. Example 11 obtains a single RGB-D sample and its camera intrinsics; a complete sequence is not required.

UniPose9D^[5] 与 Qwen^[7] 按需准备。重建资源与位姿估计使用同一个 WAPR_CACHE_DIR 根目录：源码依赖位于 cache/sources/，模型文件位于 cache/weights/，共用演示输入位于 cache/samples/。示例 11 获取单帧 RGB-D 小样及相机内参，无需下载完整序列。

python -m wapr.bootstrap --feature unipose9d qwen

Prepare the selected features with the commands on this page before running reconstruction. Use upstream models under their respective licenses. Intermediate outputs are saved in outputs/11_reconstruct_object/stages/.

运行重建前，用本页命令准备所需功能，并遵守各上游模型许可。中间结果存于 outputs/11_reconstruct_object/stages/。

SAM2.1 uses the Ultralytics sam2.1_l.pt checkpoint. SAM3D weights are obtained from ModelScope by default under the upstream model license. MoGe checkpoints remain paired with the selected pipeline.

SAM2.1 使用 Ultralytics 的 sam2.1_l.pt。SAM3D 默认从魔搭获取权重，仍须遵守上游模型许可；MoGe 权重与当前流程匹配。

Install reconstruction features安装重建扩展

On Linux, prepare SAM3D and RoMa with the command below. WAPR uses a compatible reconstruction environment while retaining the PyTorch environment used for pose estimation. Model weights are cached for later runs.

在 Linux 上运行以下命令准备 SAM3D 和 RoMa。WAPR 使用兼容的重建环境，并保留位姿估计所用的 PyTorch 环境。模型权重下载后会缓存，供后续运行复用。

python -m wapr.bootstrap --feature sam3d roma

SAM3D weights download from ModelScope. Use is subject to the upstream model license. After exporting the bundled examples, run example 11 to reconstruct an object and example 12 to use the reconstructed model.

SAM3D 权重从魔搭下载，使用须遵守上游模型许可。导出随包示例后，运行示例 11 重建物体，再运行示例 12 使用重建模型。

Robot simulation机器人仿真

Use a compatible Linux environment with Python 3.10 or newer. Robotics uses ManiSkill^[11] and SAPIEN^[12]. Prepare robotics with the command below; then export the bundled examples using export_examples as shown in the installation guide.

使用兼容的 Linux 环境及 Python 3.10 或更高版本。机器人使用 ManiSkill^[11] 与 SAPIEN^[12]。先用以下命令准备机器人功能，再按安装指南使用 export_examples 导出随包示例。

python -m wapr.bootstrap --feature robot --check
python -m wapr.bootstrap --feature robot

Use python -m wapr.bootstrap --feature robot --check to inspect the installed dependencies without starting a simulation. The first robot example then prepares its required task resources.

运行 python -m wapr.bootstrap --feature robot --check 可检查已有依赖，不启动仿真。首次运行机器人示例时会准备所需任务资源。

Example 14 needs ManiSkill's Bridge^[16] assets. Example 14 downloads them automatically; the carrot and eggplant collision meshes are stored under cache/samples/maniskill/tasks/bridge_v2_real2sim_dataset/custom/models/. An explicit MS_ASSET_DIR overrides the asset root and the example derives BRIDGE_ROOT from it.

示例 14 需要 ManiSkill 的 Bridge^[16] 资源。示例 14 会自动下载资源； cache/samples/maniskill/tasks/bridge_v2_real2sim_dataset/custom/models/ 下保存胡萝卜和茄子的碰撞网格。显式设置 MS_ASSET_DIR 时使用该资源根目录，示例会据此推导 BRIDGE_ROOT。

python -m wapr.bootstrap --feature robot --check
python -m wapr.bootstrap --feature robot

Optional performance experiments可选性能实验

The inference/renderer comparison additionally needs ONNX Runtime^[14] GPU and nvdiffrast^[8] in the same Python 3.10 environment. The recorded versions are ONNX Runtime 1.23.2 and nvdiffrast 0.3.3.1. Follow the official CUDA/cuDNN compatibility instructions and nvdiffrast installation instructions for the installed PyTorch/CUDA stack. These are optional benchmark dependencies, outside the default estimator installation.

推理与渲染后端对照还需在同一 Python 3.10 环境中准备 ONNX Runtime^[14] GPU 与 nvdiffrast^[8]，记录版本分别为 1.23.2 与 0.3.3.1。按官方 CUDA/cuDNN 兼容说明与 nvdiffrast 安装说明匹配当前 PyTorch/CUDA 栈。这些仅用于可选性能实验，不属于默认估计器依赖。